Re: [Flightgear-devel] Rembrandt performance
"* you have tree density to 0.7 in the shots, I had it at 2.4 - since there's lots of trees in the scene (it's tropical forest), that would be expected to have an impact" Yes the vegetation has a huge impact, i experienced from 0;7 to 2.5 the fps decrease is 40 %, with or without Rembrandt. The clouds density is right now less sensitive, only the visibility range as an impact ( however, because of the eye candy we are using the maximum ). Two GPU embedded should not be an issue, though our system manager did not noticed any significative improvement with the SLI architecture with FG , but it cant get some comparison since, with it, the display is supposed to be divided (AFS or SFR). It would be interesting to know how these two GPU are working together. Ahmad On 17 June 2013 08:20, Renk Thorsten wrote: > > It could help you to understand why you are getting that poor > > performances > > with your pretty fast beast GTX 670M. > > Thanks, though I remain mystified. > > There are few differences I can spot: > > * you have tree density to 0.7 in the shots, I had it at 2.4 - since > there's lots of trees in the scene (it's tropical forest), that would be > expected to have an impact > > * you have cloud density set to 0.4 whereas I have set it to 1, and also > in vaguely remember seeing a bit more clouds > > * you have dds textures used, I have used the regional set - my > understanding was that this should affect texture loading times and > available resolution, but not really runtime performance, but this needs to > be checked > > -> So I have 4 times the number of trees in the scene and 2-3 times the > number of cloud sprites. I'm not completely sure how Rembrandt manages > trees, but it could be that since they're semi-transparent they're like the > clouds taken out of the deferred rendering approach. Multiple cloud sprites > in a row are a significant drain on both vertex and fragment shaders as you > may need hundreds of texture lookups - so the available performance for > Rembrandt-specific tasks isn't quite the same. > > My gut feeling is that this can't account for a factor 4 in framerate > though, especially since there is still the pixel number working the > opposite way this time - I will have to test this. > > In addition there's the question of having two GPUs. I wonder if they're > both utilized for the job - if so, maybe one needs such a setup to get > above 30 fps with shadows? It would be helpful if a few other Rembrandt > users could give some indication of what framerates they usually get and > what the main framerate killers are. > > * Thorsten > > > > -- > This SF.net email is sponsored by Windows: > > Build for Windows Store. > > http://p.sf.net/sfu/windows-dev2dev > ___ > Flightgear-devel mailing list > Flightgear-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/flightgear-devel > -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
Re: [Flightgear-devel] Rembrandt performance
> It could help you to understand why you are getting that poor > performances > with your pretty fast beast GTX 670M. Thanks, though I remain mystified. There are few differences I can spot: * you have tree density to 0.7 in the shots, I had it at 2.4 - since there's lots of trees in the scene (it's tropical forest), that would be expected to have an impact * you have cloud density set to 0.4 whereas I have set it to 1, and also in vaguely remember seeing a bit more clouds * you have dds textures used, I have used the regional set - my understanding was that this should affect texture loading times and available resolution, but not really runtime performance, but this needs to be checked -> So I have 4 times the number of trees in the scene and 2-3 times the number of cloud sprites. I'm not completely sure how Rembrandt manages trees, but it could be that since they're semi-transparent they're like the clouds taken out of the deferred rendering approach. Multiple cloud sprites in a row are a significant drain on both vertex and fragment shaders as you may need hundreds of texture lookups - so the available performance for Rembrandt-specific tasks isn't quite the same. My gut feeling is that this can't account for a factor 4 in framerate though, especially since there is still the pixel number working the opposite way this time - I will have to test this. In addition there's the question of having two GPUs. I wonder if they're both utilized for the job - if so, maybe one needs such a setup to get above 30 fps with shadows? It would be helpful if a few other Rembrandt users could give some indication of what framerates they usually get and what the main framerate killers are. * Thorsten -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev ___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
Re: [Flightgear-devel] Rembrandt performance
Hello, You may want some screenshots which expose the performances with rembrandt at WAAJ and the PAF team DR-400 and the computer with Nvidia GPU 560 TI i refer to . It could help you to understand why you are getting that poor performances with your pretty fast beast GTX 670M. Let's consider right now the 560 TI is an old outdated GPU. https://3291185c-a-62cb3a1a-s-sites.googlegroups.com/site/grtuxhangarctd/other-download/fgfs-screen-506.jpg https://3291185c-a-62cb3a1a-s-sites.googlegroups.com/site/grtuxhangarctd/other-download/fgfs-screen-508.jpg https://3291185c-a-62cb3a1a-s-sites.googlegroups.com/site/grtuxhangarctd/other-download/fgfs-screen-510.jpg and the screenshot which expose the cpu / gpu system usage https://3291185c-a-62cb3a1a-s-sites.googlegroups.com/site/grtuxhangarctd/other-download/Flightgear-cpu-gpu_resource.jpg Thank Ahmad On 15 June 2013 13:42, grtuxhangar team wrote: > Hello, > > Just tested with an other computer 2x GPU 560 TI /SLI screen 1900x1200. > better than mine. > At WAAJ and the PAF team DR-400 > Getting an average of 50 fps: rembrandt-enabled , real weather fetch, > shaders water 5, urban 3, model 3, and multithreading ( the whole cpu > loading is only at 25 % ). > Nvidia FXAA , anisotropic filtering x16, texture sharpening. > > Since SLI does not improve the display with FG we may consider these > performances to be compared with yours. > > > > Ahmad > > > On 13 June 2013 12:09, Renk Thorsten wrote: > >> > Which screen size ? >> > With my old GPU Geforce 9600 GT at 1024x800 i never got less than 24 fps >> > (usually 30 fps). It is when using FG 2.10. >> > Decreasing on the fly the screen size increase the fps. >> >> Yeah, well, fragment shader load (and hence deferred rendering) scales >> with the number of pixels. My fullscreen is 1920x1080, which means I have >> about 2.5 times the pixel number of your screen, which means that if I >> scale your 24 fps to my screen, I would get about 9-10 fps. If I decrease >> that by the 30% you mention (since I run recent GIT), then I'd end up with >> 6 fps compared with my 15, which doesn't seem grossly out of place. >> >> But to render 2 megapixels isn't an unusual demand - that's about what >> the human eye can resolve from a normal view distance. >> >> * Thorsten >> >> -- >> This SF.net email is sponsored by Windows: >> >> Build for Windows Store. >> >> http://p.sf.net/sfu/windows-dev2dev >> ___ >> Flightgear-devel mailing list >> Flightgear-devel@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/flightgear-devel >> > > -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
Re: [Flightgear-devel] Rembrandt performance
Hello, Just tested with an other computer 2x GPU 560 TI /SLI screen 1900x1200. better than mine. At WAAJ and the PAF team DR-400 Getting an average of 50 fps: rembrandt-enabled , real weather fetch, shaders water 5, urban 3, model 3, and multithreading ( the whole cpu loading is only at 25 % ). Nvidia FXAA , anisotropic filtering x16, texture sharpening. Since SLI does not improve the display with FG we may consider these performances to be compared with yours. Ahmad On 13 June 2013 12:09, Renk Thorsten wrote: > > Which screen size ? > > With my old GPU Geforce 9600 GT at 1024x800 i never got less than 24 fps > > (usually 30 fps). It is when using FG 2.10. > > Decreasing on the fly the screen size increase the fps. > > Yeah, well, fragment shader load (and hence deferred rendering) scales > with the number of pixels. My fullscreen is 1920x1080, which means I have > about 2.5 times the pixel number of your screen, which means that if I > scale your 24 fps to my screen, I would get about 9-10 fps. If I decrease > that by the 30% you mention (since I run recent GIT), then I'd end up with > 6 fps compared with my 15, which doesn't seem grossly out of place. > > But to render 2 megapixels isn't an unusual demand - that's about what the > human eye can resolve from a normal view distance. > > * Thorsten > > -- > This SF.net email is sponsored by Windows: > > Build for Windows Store. > > http://p.sf.net/sfu/windows-dev2dev > ___ > Flightgear-devel mailing list > Flightgear-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/flightgear-devel > -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
Re: [Flightgear-devel] Rembrandt performance
> Which screen size ? > With my old GPU Geforce 9600 GT at 1024x800 i never got less than 24 fps > (usually 30 fps). It is when using FG 2.10. > Decreasing on the fly the screen size increase the fps. Yeah, well, fragment shader load (and hence deferred rendering) scales with the number of pixels. My fullscreen is 1920x1080, which means I have about 2.5 times the pixel number of your screen, which means that if I scale your 24 fps to my screen, I would get about 9-10 fps. If I decrease that by the 30% you mention (since I run recent GIT), then I'd end up with 6 fps compared with my 15, which doesn't seem grossly out of place. But to render 2 megapixels isn't an unusual demand - that's about what the human eye can resolve from a normal view distance. * Thorsten -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev ___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
Re: [Flightgear-devel] Rembrandt performance
Hello, Which screen size ? With my old GPU Geforce 9600 GT at 1024x800 i never got less than 24 fps (usually 30 fps). It is when using FG 2.10. Decreasing on the fly the screen size increase the fps. I am using Linux KDE4. However with the recent FG Git, i am losing at least 30 % of performances :( same scenery same conditions. At the moment i can't explain that "issue". BTW my rembrant parameters are : 8192 3 4 1.0 5.0 50.0 500.0 > I've had a my first short go with Rembrandt on my new machine yesterday. The > test case was a small airport in Sulawesi (Indonesia) (WAAJ) where I'm > discovering a very nice scenery. There are no static or shared models to > speak of, there is some forest around, and that's basically it. I chose > fair weather, i.e. a modest cloud cover. The aircraft was the PAF team > DR-400 in the latest version. > > All Rembrandt functions work out of the box very nicely. I started with a > dawn scene and tried the landing light illumination first. This gave me a > good 30 fps. I then switched to noon and tried shadows. I have to say that > since I am more the VFR virtual pilot, I almost never fly at night, > lightmap for internal illumination work fine for me, and so shadows are the > main selling point of Rembrandt which attracts me. > > The initial shadows coming up by default were rather ragged and flickery > (the last is a problem for me, I tend to get headache when looking at some > sort of flickers unfortunately), so I played with shadow map size, cascade > ranges and filtering till I had a nice result. To my dismay, at this point > the framerate counter gave me a mere 15 fps (no shader effects on at this > point). > > For comparison, the same scene renders in Atmospheric Light Scattering with > all details maxed out (including tree motion) with solid 60 fps. > > Am I doing anything wrong? Did I miss any optimization which makes the > shadows run fast enough? Am I just unlucky and my system has some > unspecified problems chewing Rembrandt? Does anyone else get significantly > higher framerate out of shadows with filtering? I am running on an GeForce > GTX 670M, which is usually a pretty fast beast. > > I mean, maybe it's just me, but this appears to confirm a suspicion I wrote > earlier that trying to pack ALS functionality into Rembrandt will end up > being way too slow. If I have a mere 15 fps before any shaders, then I > can't reasonably apply 800 lines of extra computations and expect no > performance impact. > > Does anyone have a semi-solid case which would argue that this would be fast > enough? I'm sort of trying to make my mind up if I should focus on that > before the next release (which is why I did the test), but it seems > hopeless to me. It's okay and flyable as it stands, but I don't see how to > cram lots of extra stuff in. > > * Thorsten GrthTeam https://sites.google.com/site/grtuxhangar -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev ___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
Re: [Flightgear-devel] Rembrandt performance
Thorsten > Sent: 13 June 2013 07:25 > To: FlightGear developers discussions > Subject: [Flightgear-devel] Rembrandt performance > > > I've had a my first short go with Rembrandt on my new machine yesterday. > The test case was a small airport in Sulawesi (Indonesia) (WAAJ) where I'm > discovering a very nice scenery. There are no static or shared models to > speak of, there is some forest around, and that's basically it. I chose fair > weather, i.e. a modest cloud cover. The aircraft was the PAF team DR-400 in > the latest version. > > All Rembrandt functions work out of the box very nicely. I started with a > dawn scene and tried the landing light illumination first. This gave me a good > 30 fps. I then switched to noon and tried shadows. I have to say that since I > am more the VFR virtual pilot, I almost never fly at night, lightmap for internal > illumination work fine for me, and so shadows are the main selling point of > Rembrandt which attracts me. > > The initial shadows coming up by default were rather ragged and flickery (the > last is a problem for me, I tend to get headache when looking at some sort of > flickers unfortunately), so I played with shadow map size, cascade ranges and > filtering till I had a nice result. To my dismay, at this point the framerate > counter gave me a mere 15 fps (no shader effects on at this point). > > For comparison, the same scene renders in Atmospheric Light Scattering with > all details maxed out (including tree motion) with solid 60 fps. > > Am I doing anything wrong? Did I miss any optimization which makes the > shadows run fast enough? Am I just unlucky and my system has some > unspecified problems chewing Rembrandt? Does anyone else get > significantly higher framerate out of shadows with filtering? I am running on > an GeForce GTX 670M, which is usually a pretty fast beast. > > I mean, maybe it's just me, but this appears to confirm a suspicion I wrote > earlier that trying to pack ALS functionality into Rembrandt will end up being > way too slow. If I have a mere 15 fps before any shaders, then I can't > reasonably apply 800 lines of extra computations and expect no performance > impact. > > Does anyone have a semi-solid case which would argue that this would be > fast enough? I'm sort of trying to make my mind up if I should focus on that > before the next release (which is why I did the test), but it seems hopeless > to me. It's okay and flyable as it stands, but I don't see how to cram lots of > extra stuff in. > I think your numbers are pretty representative. 15 fps is definitely not enough IMO. I would say that 30 fps would be a good aiming point. Smoothness is also a factor. Vivian -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev ___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel
[Flightgear-devel] Rembrandt performance
I've had a my first short go with Rembrandt on my new machine yesterday. The test case was a small airport in Sulawesi (Indonesia) (WAAJ) where I'm discovering a very nice scenery. There are no static or shared models to speak of, there is some forest around, and that's basically it. I chose fair weather, i.e. a modest cloud cover. The aircraft was the PAF team DR-400 in the latest version. All Rembrandt functions work out of the box very nicely. I started with a dawn scene and tried the landing light illumination first. This gave me a good 30 fps. I then switched to noon and tried shadows. I have to say that since I am more the VFR virtual pilot, I almost never fly at night, lightmap for internal illumination work fine for me, and so shadows are the main selling point of Rembrandt which attracts me. The initial shadows coming up by default were rather ragged and flickery (the last is a problem for me, I tend to get headache when looking at some sort of flickers unfortunately), so I played with shadow map size, cascade ranges and filtering till I had a nice result. To my dismay, at this point the framerate counter gave me a mere 15 fps (no shader effects on at this point). For comparison, the same scene renders in Atmospheric Light Scattering with all details maxed out (including tree motion) with solid 60 fps. Am I doing anything wrong? Did I miss any optimization which makes the shadows run fast enough? Am I just unlucky and my system has some unspecified problems chewing Rembrandt? Does anyone else get significantly higher framerate out of shadows with filtering? I am running on an GeForce GTX 670M, which is usually a pretty fast beast. I mean, maybe it's just me, but this appears to confirm a suspicion I wrote earlier that trying to pack ALS functionality into Rembrandt will end up being way too slow. If I have a mere 15 fps before any shaders, then I can't reasonably apply 800 lines of extra computations and expect no performance impact. Does anyone have a semi-solid case which would argue that this would be fast enough? I'm sort of trying to make my mind up if I should focus on that before the next release (which is why I did the test), but it seems hopeless to me. It's okay and flyable as it stands, but I don't see how to cram lots of extra stuff in. * Thorsten -- This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev ___ Flightgear-devel mailing list Flightgear-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/flightgear-devel